Core Concepts
LLP-Bench is the first large scale tabular LLP benchmark with diverse datasets, enabling systematic study and design of new LLP techniques.
Abstract
The content introduces LLP-Bench, a benchmark for tabular LLP datasets, addressing the lack of large-scale benchmarks. It proposes metrics to quantify dataset hardness and evaluates 9 SOTA baselines on diverse datasets from Criteo CTR and SSCL. The analysis includes performance trends, dataset classifications, and outlier explanations.
Stats
"LLP-Bench is the first large scale tabular LLP benchmark."
"LLP-Bench proposes metrics to quantify dataset hardness."
"9 SOTA baselines are evaluated on diverse datasets from Criteo CTR and SSCL."
Quotes
"LLP-Bench is the first large scale tabular LLP benchmark."
"LLP-Bench proposes metrics to quantify dataset hardness."
"9 SOTA baselines are evaluated on diverse datasets from Criteo CTR and SSCL."